The great software quality collapse or, how we normalized catastrophe

Original Article ↗ Hacker News Discussion ↗

Tradeoffs and “Good Enough” Software

  • Many argue chasing perfect software is unrealistic; real-world constraints, business survival, and time-to-market dominate.
  • Others push back: there’s a wide gap between perfection and “leaks 32GB,” and much of today’s slop reflects profit-maximization, not necessity.
  • Startups are framed as optimizing for speed, but several point out that bad code can also kill a startup by slowing iteration and burning engineers.

Has Software Quality Really Collapsed?

  • One camp sees a dramatic decline in everyday quality: constant bugs in apps, OSes, “smart” devices, and endless forced updates; users act as testers.
  • Another camp calls this nostalgia: 90s/00s systems crashed constantly, security was worse, and jank was normal; what changed is visibility and scale.
  • Some distinguish eras: early DOS/NetWare-style systems were extremely stable but simpler; later GUI/networked systems janky; now more stable but bloated and slower in UI.

Abstractions, Performance, and Physical Limits

  • Commenters debate whether growing abstraction layers inherently degrade quality or are necessary to handle complexity.
  • Some note memory/CPU trends, dead or slowing Moore’s law, power limits, and argue future energy constraints may force efficiency.
  • Others counter that energy/data-center panic is overstated relative to other global energy uses.

Incentives, Markets, and Regulation

  • Recurring theme: we get the quality that incentives pay for. Short-term profit, cheap hardware, and easy updates favor shipping features over robustness.
  • Oligopolies and moats (OSes, collaboration tools, security products) let dominant vendors ship poor quality without losing customers; insurance and compliance can even mandate flawed products.
  • Several argue only regulation, liability, or “skin in the game” will change behavior, pointing to safety-critical domains (aviation, medical) where standards are strict.

AI/LLMs and Developer Skill

  • Many see AI as “weaponizing existing incompetence”: juniors may never learn debugging or design, relying on tools that churn plausible but buggy code.
  • Others highlight AI’s value in reviewing configs, finding security issues, and assisting experts, but stress it’s not a silver bullet and produces many false positives.
  • Some predict AI could eventually help reduce bloat by reasoning about unnecessary abstractions; others are skeptical given AI is trained on today’s messy code.

Professionalism, Craft, and QA

  • Comparisons to electricians, plumbers, and doctors: past chaotic phases eventually led to standards and licensing; software is seen as still pre-standardization.
  • QA roles are perceived as shrinking, with “test in production” normalized and users bearing the cost.
  • Several IT/ops voices express frustration at cleaning up after high-paid developers and brittle, over-layered stacks.

Meta: Article Quality and AI “Slop”

  • A sizeable subthread doubts the article’s own quality, calling it formulaic or LLM-assisted, citing repeated rhetorical patterns and unverified claims (e.g., tech stacks, incidents).
  • Broader worry: AI-generated text floods discourse with polished but shallow writing, raising the cost of finding genuinely thoughtful analysis.